You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I am writing a python code to convert the temporary *.npy output files to netCDF with minimal memory overhead, so that the conversion succeeds even if the output file is larger than the computers memory. For my runs, even with the fixes in #1095, my large runs fail. I have some questions. I am working from the current development branch.
In pset_info.npy:
What is the purpose of the fields 'var_names_once', 'var_dtypes_once', and 'file_list_once'?
Is 'lonlatdepth_dtype' ever inconsistent with 'var_dtypes', and if so which should be preferred?
Finally, are the values in "id" unique to each drifter, or will they be re-used for multiple drifters? I am finding cases where a drifter id is only present in some output steps for a given rank. I.e. in directory "out-PZHRJZWC/0/" id=33 will be present in files 0.npy to 66.npy, except for 63.npy, to pick a random example. I am trying to figure out how this could be.
All runs are made with MPI, so there are multiple directories.
I will share this code when I have these issues fixed.
Thanks,
Jamie
The text was updated successfully, but these errors were encountered:
Thanks so much for taking up this challenging biut very relevant task, @JamiePringle! And sorry that I've been quiet for a few weeks and also didn't respond to the discussion in #1091; I've been inundated with teaching and marking, as this is my annual intense-teaching period.
To quickly answer your questions so you can continue your work
What is the purpose of the fields 'var_names_once', 'var_dtypes_once', and 'file_list_once'?
These are for variables that are only written once (to_write='once'), for example because they don't change during a simulation. This saves storage, as each particle only requires one item stored, instead of a vector
Is 'lonlatdepth_dtype' ever inconsistent with 'var_dtypes', and if so which should be preferred?
Not sure if they could be inconsistent (I'd hope not!) but I would say that lonlatdepth_dtype has precedence. Perhaps issue a warning if the two are not the same, so we can keep an eye on it?
Finally, are the values in "id" unique to each drifter, or will they be re-used for multiple drifters?
Hmm, @CKehl has a much deeper understanding of this; I think that for ParticleFiles the IDs are unique
Hope this helps! I'm going back to teaching but should resurface to engage with Parcels development in a week or two
I am writing a python code to convert the temporary *.npy output files to netCDF with minimal memory overhead, so that the conversion succeeds even if the output file is larger than the computers memory. For my runs, even with the fixes in #1095, my large runs fail. I have some questions. I am working from the current development branch.
In pset_info.npy:
Finally, are the values in "id" unique to each drifter, or will they be re-used for multiple drifters? I am finding cases where a drifter id is only present in some output steps for a given rank. I.e. in directory "out-PZHRJZWC/0/" id=33 will be present in files 0.npy to 66.npy, except for 63.npy, to pick a random example. I am trying to figure out how this could be.
All runs are made with MPI, so there are multiple directories.
I will share this code when I have these issues fixed.
Thanks,
Jamie
The text was updated successfully, but these errors were encountered: